Closed LiJiaqi96 closed 1 month ago
Good question! In the current version, only the smit
is added. And we have updated the instruction data here.
Thanks for your reply! I noticed that some datasets like image caption -- coco has been reduced to a smaller size (100k). I'm curious about the reason and whether it is because the length of captions are relatively short? Thanks
Yes. It's relatively short and similar to the Stage2 data. Besides, in our experiments, removing most of these data does not affect the results.
OK, it's great to train faster with the same performance :)
Very happy to hear that you have updated the model with mistral LLM. Is there any place to find the newly added datasets? Thanks!